anxbbq / vecf

Code for "Personalized Fashion Recommendation with Visual Explanations based on Multi-model Attention Network"

Visually Explainable Collaborative Filtering

Requirements

• Python 2.7
• TensorFlow 1.8
• numpy 1.14

Project Structure

.
├── conf                    # Config files
├── data
    ├── categoried_data     # Experiment data 
├── results                 # Results saving
├── model          
    ├── our_model
        ├── VECFMain.py     # The main file of VECF
        ├── VECFModel.py    # The graph of VECF
        ├── FromMemory.py   # DataLoader          

Parameter Settings

Example:

data_type = 'Men/0' ######## need change ##########

batch_size = 256

global_dimension = 10

item_dimension = 10 # should be the same as global_dimension when using mul or add merging methods

hidden_dimension = 100

context_dimension = 50

word_embedding_dimension = 64

reg = 0.01

learning_rate = 0.01

optimization = 'Adam'

max_epoch = 15

K = 10

beta = 0.01

Usage

1. Install all the required packages

2. Change the path to ./model/our_model/

3. Run VECFMain.py

Dataset for evaluating visual explainations

Upon request